Welding robot welding seam recognition method based on deep learning

Abstract

The invention discloses a welding robot welding seam recognition method based on deep learning. The welding robot welding seam recognition method comprises the steps of acquiring a sample image set ofwelding seams; establishing a model of a convolution-deconvolution neural network wherein the model comprises a convolutional neural network and a deconvolutional neural network, extracting image features of a to-be-identified welding seam through the convolutional neural network, and obtaining a semantic expression of the to-be-identified welding seam through the deconvolutional neural network;wherein the convolutional neural network comprises a convolutional layer with a convolutional kernel, and offset variables are set at sampling point positions of the convolutional kernel, so that adaptive change of the sampling points of the convolutional kernel according to characteristics of a to-be-identified welding seam is realized; using the sample image set training model, acquiring the convolution-deconvolution neural network; inputting an acquired image of the weld to be identified into the convolution-deconvolution neural network, and acquiring welding types of a segmented picture and a segmented picture corresponding to the welding seams to be identified. The invention can greatly improve accuracy and efficiency of welding seam identification.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a welding seam recognition method of a welding robot based on deep learning. Background technique [0002] At present, the production and manufacture of large-scale equipment such as ships, containers, and marine equipment requires a huge amount of welding work. The structures that make up these welded components are often complex, and the thickness, size, and length of the welding materials vary, resulting in different sizes, Welds of different sizes, shapes and positions in space. In the process of identification, the environment is complex, harsh and changeable, which brings certain challenges to weld identification. [0003] In the process of welding, different welding seam types correspond to different welding parameter settings. Wrong welding parameter settings can easily lead to welding defects, thereby affecting the entire production. The traditional welding pr...

AI Technical Summary

Problems solved by technology

At the same time, the existing welding seam recognition technology mostly uses structured light and laser for auxiliary detection, or has high requirements for the texture differentiation of welded seam plates, and the recognition rate and recognition speed are not high in complex industrial environments.

Images